Objective: This study aims to uncover subtypes of Parkinson’s disease (PD) through unsupervised machine learning (ML), based on the predicted evolution of non-motor symptoms (NMS).

Background: PD displays heterogeneity in the combinations and severity of NMS, which remains poorly understood. ML can discern patterns of NMS evolution from clinical measures, allowing the identification of subtypes among PD patients. The objective of this study was to uncover such subtypes through an unsupervised ML algorithm.

Method: The SuStaIn algorithm was employed on a dataset including 956 PD subjects from the Canadian Open Parkinson Network, who were assessed using the MDS-UPDRS scale. We implemented this ML approach for the subtyping and staging of patients based on cross-sectional data (non-motor and motor scores). Following the unsupervised classification by SuStaIn, multinomial and ordinal logistic regressions were conducted to identify differences in symptoms among the uncovered subtypes, as well as differences in their predicted trajectories.

Results: 879 PD patients (33.74% females) were successfully classified into subtypes and stages. Median age was 67 years (IQR 60-73), and duration of the disease was 6.6 years (IQR 4.2-10.6). Three distinct subtypes were uncovered among participants: 291 (33.11%) corresponded to a mild motor-predominant subtype (mainly gait freezing and limitations when arising from beds or chairs), 379 (43.12%) to an apathetic subtype, and 209 (23.78%) to a cognitive subtype. There were no significant differences in MDS-UPDRS part III scores by subtype (p>0.050) but, when compared to the motor subgroup, the apathetic subtype exhibited higher scores in parts I (p=0.019) and IV (p=0.023). Furthermore, model-predicted staging was significantly associated with PD duration and Hoehn and Yahr stage (p=0.000). Analysis of staging showed higher slopes, indicating faster evolution, for speech problems (motor subtype), hygiene difficulties (apathetic subtype), and cognitive impairment and hallucinations (p<0.050) (cognitive subtype).

Conclusion: SuStaIn proves valuable in identifying potential subtypes of individuals with PD according to NMS in a large multi-center sample. Subsequent analyses should focus on validating these results in other cohorts and investigating their biological underpinnings.

References: Young, A. L., Marinescu, R. V., Oxtoby, N. P., Bocchetta, M., Yong, K., Firth, N. C., Cash, D. M., Thomas, D. L., Dick, K. M., Cardoso, J., van Swieten, J., Borroni, B., Galimberti, D., Masellis, M., Tartaglia, M. C., Rowe, J. B., Graff, C., Tagliavini, F., Frisoni, G. B., … Alexander, D. C. (2018). Uncovering the heterogeneity and temporal complexity of neurodegenerative diseases with Subtype and Stage Inference. Nature Communications, 9(1), Article 1. https://doi.org/10.1038/s41467-018-05892-0

Young, A. L., Vogel, J. W., Aksman, L. M., Wijeratne, P. A., Eshaghi, A., Oxtoby, N. P., Williams, S. C. R., Alexander, D. C., & Initiative, for the A. D. N. (2021). Ordinal SuStaIn: Subtype and Stage Inference for Clinical Scores, Visual Ratings, and Other Ordinal Data. Frontiers in Artificial Intelligence, 4. https://doi.org/10.3389/frai.2021.613261

« Back to 2024 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/uncovering-non-motor-subtypes-in-parkinsons-disease-through-machine-learning/